SK hynix Ships 12-Layer HBM4E Samples Ahead of Schedule at 16 Gbps, Tightening the AI Memory Race With Samsung
SK hynix shipped 12-high HBM4E samples on June 18, months ahead of its own second-half timeline, three weeks after Samsung's May 29 shipment.
Overview
SK hynix has begun shipping samples of its 12-layer next-generation HBM4E memory, the company announced on June 18. The move arrives months earlier than the company’s own guidance: during its first-quarter earnings call in April, SK hynix had said it planned to provide HBM4E samples in the second half of the year, but instead shipped ahead of that schedule, according to The Elec. The early sampling tightens a close race with Samsung Electronics, which sent its own 12-high HBM4E samples on May 29.
HBM4E is the enhanced, mid-generation step between HBM4 — which entered mass production earlier this year — and the HBM5 generation still on the drawing board. With AI accelerators bottlenecked more by memory bandwidth than by raw compute, the speed at which suppliers can qualify each new stack has become a competitive front in its own right.
What We Know
SK hynix’s 12-layer HBM4E reaches a maximum data processing speed of 16 Gbps per pin and improves power efficiency by more than 20 percent over the previous generation, according to SK hynix. A 12-high stack delivers 48 GB of capacity, up from 36 GB for HBM4’s 12-high configuration, The Elec reported. The same report frames the 16 Gbps figure as an improvement of as much as 45 percent over the 11 Gbps to 13 Gbps range of previous products.
The performance jump rests on a process change. SK hynix adopted its sixth-generation 10-nanometer-class 1c DRAM for HBM4E, transitioning from the fifth-generation 1b node used in HBM4, as reported by TrendForce. The Elec notes this is the first use of the 1c technology in an HBM product. To stack twelve dies while managing heat, SK hynix uses its Advanced MR-MUF packaging, which the company says improves heat resistance by 17 percent compared to the preceding HBM4, according to SK hynix.
“SK hynix has laid the foundation to strengthen its AI leadership with HBM4E based on its market-leading technological capabilities and manufacturing expertise,” said Ahn Hyun, President and Chief Development Officer, in the company’s statement.
Samsung reached the same 12-layer milestone first. Its 12-high HBM4E samples, shipped May 29, offer a stable pin speed of 14 Gbps scalable up to 16 Gbps and up to 3.6 TB/s of bandwidth per stack in a 48 GB package, according to SDxCentral. Samsung says the part improves energy efficiency by 16 percent and thermal resistance by more than 14 percent over the previous generation, and builds its base die on a 4-nanometer foundry process. The Korea Herald reports Samsung plans to extend the lineup with 32 GB eight-layer and 64 GB 16-layer variants.
“Through our advanced manufacturing capabilities and preemptive infrastructure investments, we will continue to drive the growth of the global AI memory market,” said Sang Joon Hwang, EVP and head of memory development at Samsung, according to SDxCentral.
The two announcements land against a backdrop in which Samsung began volume shipments of HBM4 in February, per TrendForce.
What We Don’t Know
Neither company has committed to a firm mass-production date for 12-layer HBM4E. SK hynix said it will work with partners for mass production “in a timely manner” without specifying when, according to its statement. Customer names and the specific AI accelerators that will adopt the new memory have not been disclosed by either supplier. It also remains unclear how the two parts will be positioned against each other commercially, given they report similar headline speeds but different efficiency and thermal figures.
Analysis
The compressed timeline is the story. SK hynix moving from a second-half guidance to a June shipment, just three weeks behind Samsung, underscores how the gap between the two leading HBM suppliers has narrowed to weeks rather than quarters. Both have now demonstrated the same 12-layer, 48 GB, 16 Gbps-class configuration on sixth-generation 1c DRAM, leaving qualification speed, packaging yield, and customer relationships — rather than headline specifications — as the likely differentiators in the next round of AI memory contracts.